Improve recognition of select-based period induction.
Rationale:
Similar to the previous CL, this helps to eliminate more dead induction.
Now, CaffeineLogic, when compiled with dx (rather than jack) improves
by a 1.5 speedup (9000us -> 6000us).
Note:
We need to run the simplifier before induction analysis to trigger
the select simplification first. Although a bit of a compile-time hit,
it seems a good idea to run a simplifier here again anyway.
Test: test-art-host
Change-Id: I93b91ca40a4d64385c64393028e8d213f0c904a8
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 6ba0963..532ea43 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -743,8 +743,10 @@
HLoopOptimization* loop = new (arena) HLoopOptimization(graph, induction);
HSharpening* sharpening = new (arena) HSharpening(graph, codegen, dex_compilation_unit, driver);
InstructionSimplifier* simplify2 = new (arena) InstructionSimplifier(
- graph, stats, "instruction_simplifier$after_bce");
+ graph, stats, "instruction_simplifier$after_inlining");
InstructionSimplifier* simplify3 = new (arena) InstructionSimplifier(
+ graph, stats, "instruction_simplifier$after_bce");
+ InstructionSimplifier* simplify4 = new (arena) InstructionSimplifier(
graph, stats, "instruction_simplifier$before_codegen");
IntrinsicsRecognizer* intrinsics = new (arena) IntrinsicsRecognizer(graph, stats);
@@ -764,6 +766,7 @@
// redundant suspend checks to recognize empty blocks.
select_generator,
fold2, // TODO: if we don't inline we can also skip fold2.
+ simplify2,
side_effects,
gvn,
licm,
@@ -771,13 +774,13 @@
bce,
loop,
fold3, // evaluates code generated by dynamic bce
- simplify2,
+ simplify3,
lse,
dce2,
// The codegen has a few assumptions that only the instruction simplifier
// can satisfy. For example, the code generator does not expect to see a
// HTypeConversion from a type to the same type.
- simplify3,
+ simplify4,
};
RunOptimizations(optimizations2, arraysize(optimizations2), pass_observer);